Can Too Much Similarity Between Donors Crowd out Charitable Donations? an Experimental Investigation of the Role of Similarity in Social Influence on Giving Behavior

Home , Charity (virtue)

Can too much similarity between donors crowd out charitable donations? An experimental investigation of the role of similarity in social influence on giving behavior

Yuan Tian, Doctoral Fellows Program 2017 Yuan Tian, Ph.D., is an Assistant Research Professor at the School of International and Public Affairs at Shanghai Jiao Tong University. Her research areas are philanthropy, nonprofit fundraising, nonprofit accountability, and social entrepreneurship. She can be reached at [email protected].

Research Question Does too much similarity between actual and potential donors actually decrease the chances that a potential donor will give or decrease the amount they will give?

Brief Abstract Previous research suggests that people are motivated to donate to a cause when other people who are similar to them have generously donated to this cause. This study examines whether there is such a thing as too much visual similarity between generous donors and potential donors, and if there is a level at which potential donors become less likely to give or give a smaller amount. It finds that while people are more likely to donate (or donate more) when other generous donors appear moderately similar to themselves, they are less likely to donate (or are likely to donate less) when other generous donors are highly similar to themselves. This suggests that organizations seeking donations should aim to emphasize moderate, but not high, facial similarity between generous donors and potential donors.

Key Findings Opportunities for Action ► Potential donors are more likely to give ► Nonprofits’ fundraising strategies should (and give a larger amount) when others take into account the fact that moderate who look moderately similar to them give similarity of appearance between generous generously. donors and potential donors works best to ► Potential donors are less likely to give (or promote giving. are likely to give less) when others who ► Emphasizing a high amount of similarity look highly similar to them give may not always be successful. generously.

Find This Paper This paper is published in Current psychology. DOI: 10.1007/s12144-020-00616-4

Research Overview Doctoral Fellows Program Current Psychology https://doi.org/10.1007/s12144-020-00616-4

Can too much similarity between donors crowd out charitable donations? an experimental investigation of the role of similarity in social influence on giving behavior

Yuan Tian1 & Sara Konrath2

# Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract How is charitable giving influenced by other donors’ giving? Do people give more in the presence of donors who are similar to themselves? Most research suggests that individuals are positively influenced by similar others across a variety of behaviors. In the charitable giving context, if similar others donate, individuals are more likely to donate (or donate more) to the same cause. Yet, prior research has paid little attention to a potential non-linear relationship between similarity and charitable giving. Is there such thing as too much similarity? A between-subject laboratory experiment (N = 140) was designed to test a non-linear relationship between individuals’ similarity to other donors and their charitable giving. The study concludes that moderate similarity promotes more giving, yet too much similarity between donors may actually crowd out charitable giving.

Keywords Social influence . Philanthropy . Charitable giving . Charitable donations . Similarity . Facial morphing

Introduction Kelman 1958). Specifically, we examine whether charitable donations are driven by one’s similarity to other donors. In 2018, individual Americans donated $292.09 billion, which accounted for 68% of the total giving amount in the United States (Giving USA 2019). Given this, it is important to un- How Similarity Influences Behavior derstand factors that influence people’s decision making in charitable giving. Why do people give their money to others? Similarity is an important psychological construct, and usually Research has identified many factors that affect charitable positively affects many kinds of individual behaviors, such as giving, including other-oriented motives (i.e. altruism, trust) aggression (Baron 1971), counter-aggression (aggressive re- and self-oriented motives (i.e. egoism, tax breaks; Konrath actions to others’ aggression) (Hendrick and Taylor 1971), and Handy 2018). The current paper focuses on the role of interpersonal attraction (Byrne 1961; Secord and Backman social influence, when people’s decision-making and behav- 1964), consumer purchasing behavior (Woodside and iors are influenced by others (Cialdini and Goldstein 2004; Davenport 1974), alcohol consumption (Andsager et al. 2006), and compliance (Burger et al. 2004; Silvia 2005). In general, similarity also affects a variety of prosocial behaviors in a positive way, such as helping (Sole et al. 1975), Electronic supplementary material The online version of this article comforting and rescuing (Eagly and Koenig 2006), coopera- (https://doi.org/10.1007/s12144-020-00616-4) contains supplementary tion (Sinervo and Clobert 2003), and positive social interac- material, which is available to authorized users. tions (Boivin et al. 1995; Lee et al. 1999). * Yuan Tian [email protected] Similarity in Charitable Giving

1 School of International and Public Affairs, Shanghai Jiao Tong University, No. 1954 Huashan Rd. Xin Jian Building, Room 261, Similarity also plays an important role in charitable giving, Xuhui District, Shanghai 200040, CN, China specifically. With charitable giving, there are different ways 2 Lilly Family School of Philanthropy, Indiana University, that similarity can affect donation behavior because of the Indianapolis, IN, USA different roles involved. Curr Psychol

Donor Similarity to Solicitors People are more likely to give scored high in narcissism. For less narcissistic people, simi- (and give more) when they are asked by similar others (or larity appeared to backfire, with higher similarity leading to asked in the presence of similar others), for example, by those more aggression (see Fig. 2 in their paper). Another study who share the same religious background or first name measured different levels of similarity using participants’ (Bekkers 2010; Yinon and Sharon 1985). self-report rating of another person’s photograph. It found that participants reported lower attraction when similarity was too Donor Similarity to Recipients In addition, individuals are high (Penton-Voak et al. 1999). Another study used facial more likely to give (and give more) to similar recipients, for morphing to manipulate similarity between voters and politi- example, to those who share the same religious beliefs (Helms cal candidates. It found that among females, but not males, and Thornton 2012; Yinon and Sharon 1985) or to other in- high similarity led to lower warm feelings, voting intentions, group recipients who share the same political views, sports- and attractiveness ratings of the candidates (Bailenson et al. team preferences, or music preferences (Ben-Ner et al. 2009). 2009). Yet, some research has found a positive effect of increasing Donor Similarity to Other Donors In the current paper, we levels of similarity. For example, one study manipulated four examine how individuals’ responses differ depending upon levels of similarity that participants shared with other people their level of facial similarity to other donors. Studies have (i.e., 2, 4, 6, or 8 shared activity preferences). The greater found that people are influenced when they see similar others number of activities that participants believed they shared making donations. For example, a field experiment on tourist with others (i.e. higher similarity), the more attractive those skiers found that 44% of participants made a charitable dona- others were rated (Moreland and Zajonc 1982). Another study tion when they were told that other fellow skiers made a do- found that participants in an uncommon fingerprint condition nation, while only 22% of participants donated when partici- (high similarity) were more likely to comply with the request- pants did not have any information about previous donors’ er than participants in the common fingerprint (moderate sim- charitable giving (Heldt 2005). Another field experiment on ilarity) or control conditions (Burger et al. 2004). Another a sample of National Public Radio (NPR) members found that study used facial morphing to manipulate high similarity, participants donated more after being told that another NPR and found that high similarity increased participants’ cooper- member of the same gender donated, than when they were told ation in public good games compared with no morphing (i.e. that another NPR member of the other gender donated (Shang low similarity) (Krupp et al. 2008). et al. 2007). Thus, at times, people may be more likely to Taken together, because there are very few studies donate in the presence of a similar other. that manipulate more than one level of similarity, it is difficult to know if there is such thing as too much Research Question and Hypothesis Development similarity. In addition, no studies that we are aware of have examined potential negative effects of high simi- We aim to extend previous work on the effect of facial similarity within charitable giving contexts. In this study, we larity to other donors in charitable giving. As reported above, manipulate different levels of donor-donor similarity similarity can affect donation behavior, yet, these studies only using facial morphing software, so that we can examine manipulate a single level of facial similarity. To our knowl- how different degrees of facial similarity between do- edge, studies rarely investigate the effects of multiple levels of nors affect giving behavior. facial similarity on behaviors (i.e. only in the context of Based on the theories outlined below, we specifically hy- political voting; see Bailenson et al. 2009) and no study has pothesize that other donors who are moderately facially sim- directly manipulated different levels of facial similarity to in- ilar to participants will increase their donation behavior, but vestigate its effects on charitable donations (Tian and Konrath other donors that are highly facially similar to participants will 2019). This is important to do because one should not assume actually cause donation behavior to decrease. that just because moderate levels of similarity appear to be Similarity Attraction Theory posits that interpersonal at- beneficial, that higher levels of facial similarity will also be traction increases as similarity increases. In Byrne’soriginal beneficial. (Byrne 1961) paper, he manipulated similarity through atti- Scholars have only rarely manipulated multiple levels of tude sharing and found that participants liked a stranger better similarity in domains other than charitable giving, with incon- if the stranger had similar attitudes compared to dissimilar sistent effects. For example, one study found a potential attitudes. This positive relationship between similarity and backfiring effect of too much similarity (Konrath et al. attraction has been tested and confirmed in a variety of con- 2006). The study manipulated similarity by telling participants texts, including small groups (Lott and Lott 1965), applicant- that they shared either a common fingerprint type (moderate employee fit (Van Hoye and Turban 2015), friendship similarity) or a rare one (high similarity). Increasing levels of (Mitteness et al. 2016), and supervisor-subordinate relation- similarity led to lower aggression, but only among those who ships (Bakar and McCann 2014). Curr Psychol

However, research has found some mixed effects of simi- Brief Overview of the Current Study larity; that is, interpersonal similarity could have either positive or negative effects on attraction (Penton-Voak et al. 1999). In a college student sample, we used facial morphing technol- Additionally, similarity could also have a curvilinear effect in ogy to manipulate three different levels of self-other similari- intergroup relationships (Jetten et al. 1998). ty: Low, Moderate, and High. We then gave participants an Optimal Distinctiveness Theory can help to explain these opportunity to make a real charitable donation. To our knowl- results. It posits that individuals have two fundamental yet com- edge, this is the first time that researchers have tested the peting needs: the need for assimilation and the need for effects of three different levels of facial similarity on prosocial differentiation. Brewer (1991) argues that individuals constantly behavior. (See Fig. 2 for study flow.) adjust the levels of both needs based upon feelings of belonging- The current study is important for both theory and real life. ness to a group. When these feelings are high, their need for The major contribution of the study is to investigate a curvi- assimilation decreases, and their need for differentiation in- linear relationship between facial similarity and charitable be- creases. But when these feelings are low, their need for assimi- havior, which can help us to better understand social influence lation increases, and their need for differentiation decreases. processes in charitable giving. Practically, most people likely Simply speaking, individuals try to reach the most optimal, or believe that similarity to other donors would increase dona- comfortable, condition by balancing these two competing needs. tions, and that the more similarity, the better. However, is there Applying Similarity Attraction Theory to charitable giving, a certain amount of similarity that is too much? This study we posit that individuals are more likely to give (or give more) investigates the effect of facial similarity at different levels when moderately similar others give, compared to when less sim- including a high level of similarity that could be too much, ilar others give (or no information is given). This is because mod- and could backfire in charitable giving. erately similar others are seen as more attractive to them. Yet, what if the self-other similarity becomes too high? Research has found initial evidence of the potential negative effects of similarity in Method intergroup contexts (Jetten et al. 1998) and interpersonal relationships (Penton-Voak et al. 1999), and it is possible that the effects of Participants and Design similarity in charitable giving could be non-linear as well. Based on Optimal Distinctiveness Theory, the need for assimilation and The current study recruited 140 fluent English-speaking stu- the need for differentiation are continually in tension. In the char- dents over the age of 18 from a Midwestern university campus itable giving context, people might be less likely to donate when a for a four group experimental study conducted between highly similar other donates, in order to differentiate themselves March 2016 and July 2017 (Mean age = 22.6, SD = 5.4). from that person. The final sample (N = 140) was 76% female, with an ethnic Thus overall, this paper hypothesizes curvilinear effects of distribution of: 56% Caucasian, 24% Asian, 18% African- self-other facial similarity between donors on individuals’ American, and 2% Hispanic/Latino. charitable giving (See Fig. 1). We expect to find a higher The sample size was determined prior to data collection likelihood of giving (and larger amounts) from individuals and data analyses were conducted after the completion of data when moderately similar donors give generously to charity, collection. The study was approved by the local Institutional whereas a lower likelihood of giving (and smaller amounts) Review Board (IRB). All procedures performed in the study when highly similar donors give generously. involving human participants were in accordance with the ethical standards of the local IRB.

Power Analysis

We used G*Power 3.1 software to conduct a power analysis at α = .05 (two-tailed) with a power of .80. The results indicated that our sample size of 140 was sufficient to detect a small effect size of Cohen’sd=.0566(η2 =.07)orlargerwith80% power for a simple design of four conditions.

Procedures

Cover Story Participants were told that this study examined online social interactions, and that in order to protect their Fig. 1 Study hypotheses confidentiality, experimenters scheduled participants at Curr Psychol

Fig. 2 Study flow different individual sessions. Experimenters told them that whichwas80%ofthestudypayment.IntheModerate they would take a photo, and then they might be paired with Similarity condition, participants saw a photograph that a participant from a previous session for an online social in- consisted of 25% of themselves and 75% of the hypothetical teraction, and that their photo might be used for future ses- donor. They were also given information about his/her dona- sions on a random basis. During debriefing, all participants tion amount (80% of study payment). In the High Similarity reported believing the cover story. condition, participants saw a photograph that consisted of Participants were paired with a hypothetical other person to 49% of themselves and 51% of the hypothetical donor. They control for confounding variables, by matching participants were also given information about his/her donation amount and the other person on gender, ethnicity, age, hair style, and (80% of study payment). (neutral) facial expression. Eight standard photographs were Figure 3 shows two sets of examples using two research taken from the Chicago Faces Database to represent four ma- assistants’ photos on the right column to represent partici- jor ethnic groups (Asian, Black, White, and Latino) in both pants’ original photos. The hypothetical standard donors’ genders. These eight standard photographs were converted photos on the left column represent the unmorphed photos into US visa photo size and stored in the lab computer for of the other donor that participants would see on the computer manipulating different levels of self-other facial similarity. screen in the Low Similarity condition. The middle two col- umns represent morphing examples in the Moderate Self-Other Facial Similarity Manipulation This study borrowed Similarity condition (25% from the research assistants and the similarity manipulation from a prior study finding that 75% from the hypothetical donor) and in the High Similarity individuals are more likely to vote for a political candidate condition (49% from the research assistants and 51% from the whose face appears similar to their own (Bailenson et al. hypothetical donor) respectively.1 2009). This experiment adopted similar manipulation methods using the facial morphing software, Magic Morph, Morphing Procedures and Pre-Survey After providing con- to morph a participant’s photograph with a hypothetical do- sent, participants’ color photographs were taken using a digi- nor’s photograph in the following combinations, to which tal camera and ensuring that all participants had a standard participants were randomly assigned (See Table 1). presentation (i.e. no facial hair, hair tied back, and neutral Specifically, in the No Information (control) condition, partic- facial expression). While the researcher was morphing participants completed the donation task without any information ipants’ photographs in a back room, participants completed a about another donor. In the Low Similarity condition, partici- paper-based pre-survey with a number of measures. pants saw an unmorphed photograph of a hypothetical donor The morphing steps were as follows. First, the researcher and were given information about his/her donation amount selected a standard hypothetical donor’s photograph, matched

Table 1 Experimental conditions in different morphing combinations

Conditions Facial features % from matched other “donor” Facial features % from participant

No Information Condition (n = 36) No information about another donor No information about another donor Low Similarity Condition (n = 35) 100 other donor 0 self Moderate Similarity Condition (n = 38) 75 other donor 25 self High Similarity Condition (n = 31) 51 other donor 49 self Curr Psychol

Fig. 3 Morphing examples

Matched “donor” Participant to participants’ gender and ethnicity. Next, the researcher Society3), which provides cancer patients with taxi rides resized the participant’s photo into a standard visa photo size to attend their cancer treatment appointments. at a free online photo generator website (https://www. Participants were also asked to rate different items so that persofoto.com/upload/visa-photo). Then, the researcher used the following variables could be measured. First, for Self- the morphing software, Magic Morph, to morph the Other Similarity, participants were asked to rate their facial participant’s photo with a hypothetical donor’s photo in similarity with the hypothetical donor they saw on the screen three different combinations for the three similarity on a Likert scale ranging from 1 (“not at all”)to7(“extreme- conditions: the Low Similarity, the Moderate Similarity, and ly”). Then, participants were asked to select their sense of self- the High Similarity. After the morphing was complete, the other overlap by choosing one figure from seven in which two researcher uploaded the morphed photographs to the circles overlapped at different degrees from no overlap to ex- Qualtrics survey program, which randomly assigned tremely high overlap (Aron et al. 1991). These items were participants to one of the four conditions. Thus, researchers averaged into a single measure (Cronbach α =.64). Second, were blind to the experimental condition. Only participants for Liking and Attraction, participants were asked to rate their assigned into one of the three similarity conditions saw a liking and attraction of the hypothetical donor using a Likert hypothetical donor’sphoto. scale from 1 (“not at all”)to7(“extremely”), which were The pre-survey included the following two key measures, averaged into a single score (Cronbach α =.57). Finally, par- along with some filler measures that allowed the researcher ticipants were asked to rate the impact of their donations to the enough time to morph the photograph. The 18 item local program and to the community using a Likert scale from Prosopagnosia Index assessed participants’ self-reported ability 1(“not important at all”)to7(“extremely important”). These to recognize faces (e.g. “My face recognition ability is worse were averaged into a single item (Cronbach α =.89). than most people”;Cronbachα =.96(Shahetal.2015). The 10 Next, participants were asked whether they would like to item Marlowe-Crowne Social Desirability Scale assessed par- make a donation today as well. Participants were given an ticipants’ concerns about looking good using a true-false inven- option to make a donation to the same program as the hypo- tory (e.g. “I never hesitate to go out of my way to help someone thetical donor by entering a pledge in the online survey that in trouble”;(StrahanandGerbasi1972). could range from $0 to the full study payment amount, in 25- cent increments. Online Charitable Donation Decision Participants next completed a computer-based charitable donation task, in Post-Survey The paper-based post-survey contained questions which they were randomly assigned to one of the ex- regarding participants’ previous familiarity with the American perimental conditions. Except for those in the control Cancer Society,theRoad for Recovery Program, and their condition, participants saw a hypothetical donor’spho- past experiences with cancer patients as well as some demo- tograph (named “Alex”) with the information that Alex graphic questions (i.e. age, ethnicity, and gender). (who was gender and ethnicity matched to them) had Participants’ Previous Familiarity was measured by sum- donated 80% of his/her study payment2 to the Road ming up 13 different items on a binary choice “yes” or “no”, for Recovery (local program of the American Cancer (i.e. “Have you ever heard of the American Cancer Society,” Curr Psychol

“Have you ever made a donation to the American Cancer highly correlated with each other r(140) = .48, p<.001,which Society,” or “Is there someone who is close to you who had was expected. We ran separate analyses of the condition effect cancer?”) In addition, participants also reported their frequen- on these two measures. Additionally, Liking and Attraction was cy of hearing about the American Cancer Society and the highly correlated with the manipulation check variable-Self- Road for Recovery before participating in the study (1 = very Other Similarity, r(104) = .43, p < .001. Additionally, Beliefs few times or never, 5 = several times a day). We standardized about Donation Impact was highly correlated with both dona- and combined the binary measures and the frequency mea- tion measures: donation yes/no, r(139) = .28, p<.001,and % sures into a single familiarity item (Cronbach α =.79). donated,r(139)=.41,p < .001. Thus, we conducted additional ANOVAs to investigate the condition effects on Liking and Actual Giving Behavior Finally, participants received their full Attraction and Belief about Donation Impact. payment in an envelope, in a combination of four quarters and the rest one dollar bills. In order to assure participants that the donation was voluntary and anonymous, the researcher asked Manipulation Check participants to leave whatever amount they pledged in the envelope and to put the envelope back in a black donation In order to ensure that our manipulation through facial box (even if the envelope was empty). The black donation morphing was effective, we ran an ANOVA on Self-Other box was placed on one side of the lab and the researcher sat Similarity, F(2,103) = 7.53, p <.001,η2 = .13, 95% CI [0.03, behind a wall and was unable to see the donation process. 0.24]: High Similarity: M = 4.19 SD = 1.40, n =31;Moderate Participants were told that another researcher would come Similarity: M =3.47,SD=1.02, n = 38; and Low Similarity: later to collect all the donation envelopes and the current re- M =3.06,SD=1.18, n = 35. Post-hoc pairwise comparisons searcher would not know whether the participants donated or indicated that participants in the High Similarity condition felt how much they left in the envelope. 100% of participants’ more similar to the hypothetical donor that they saw on the donations were donated to the American Cancer Society at computer screen than participants in other two conditions the end of the study. (High VS Low, t(65) = 3.85, p < .001, 95% CI [0.55, 1.72]; In our study, we measured the variables for the manipula- High VS Moderate, t(68) = 2.49, p = .02, 95% CI [0.15, tion check (Perceived Facial Similarity and Self-Other 1.30]). Even though the average Self-Other Similarity was Overlap) and variables that we expected to be mediators higher in the Moderate condition than in the Low Similarity (Liking, Attraction, and Beliefs of the Donation Impact) be- condition, these conditions were not significantly different fore measuring the online donation pledge and the actual do- from each other, t(72) =1.49, p = .14, 95% CI [−0.14, 0.97] nation behavior because such an order would allow us to ex- (See Fig. 4). amine whether our manipulation of similarity is successful and what the potential underlying process of the similarity effects on donation behavior is. Randomization Check

Results The randomization process was first checked to ensure its effectiveness by confirming that no statistical significant dif- Descriptive Statistics ferences existed across conditions.

Among the 140 participants, 102 (73%) donated and 38 (27%) Prosopagnosia Index This index measures participants’ facial did not. Among donors, the average donation was 29% of recognition (Cronbach α = .92). This ability is only relevant to their study earnings (SD = .28). There were no gender differ- the three conditions where participants saw the other donor, ences in the decision to donate, χ(1,140) = 0.30, p =.59, thus, our analyses focus on these three conditions. There were Males = 77%, Females = 72%. There were also no differences no differences between these three conditions in facial recog- by ethnicity, χ(3,140) = 1.56, p = .67, Caucasian = 73%, nition, F(2,103) = 0.20, p =.82, η2 = .004, 95% CI [0.00, Asian = 76%, African-American = 68%, and Hispanic/ 0.13]. Latino = 100%. Social Desirability Scale A check was conducted to investigate Correlations the differences in participants’ desire to look good (Cronbach α = .50) across all four conditions (three similarity conditions We report the correlations between dependent measures in and the control condition). There were no differences across Table 2. The results indicated that two measures of our key conditions in social desirability, F(3,139) = 0.41, p =.75, dependent variables, donated yes/no and % donated, were η2 = .01, 95% CI [0.00, 0.04]. Curr Psychol

Table 2 Correlations Measures 1 2 3 4 5 6 7 8

1.Donation (yes/no) 1 2.Donation (%) .48*** 1 3. Self-other Similarity −.07 .03 1 4.Prosopagnosia Index .11 .13 −.12 1 5.Social Desirability Scale .20* .06 .02 −.33 1 6.Previous Familiarity −.13 −.02 −.04 −.11 .02 1 7.Liking and Attraction .22* .17 .43*** −.02 .15 .03 1 8.Belief about Impact .28*** .41*** −.08 −.07 .15~ .08 .07 1

~(p < .10), *(p < .05), **(p < .01) and ***(p <.001)

Previous Familiarity This measure captured participants’ pre- compared to participants in all other conditions. Thus, a dum- vious familiarity with the recipient nonprofit organization and my variable was created that compared Moderate Similarity to the program in this study (the American Cancer Society and all other similarity conditions combined, and found that there the Road for Recovery) as well as their previous experiences was indeed a statistically significant result, χ(1,140) = 5.16, with cancer patients (Cronbach α = .79). A check was con- p = .02, odds ratio = 6.60, 95% CI [2.58, 16.91]. In order to ducted to investigate the differences of participants’ self- examine all six pairs of comparisons on any two conditions, reported previous familiarity across all four conditions. we used a binomial logit model. The pairwise comparisons There were no differences across the four conditions in previ- based upon the binomial logit regression indicated that the ous familiarity, F(3,139) = 1.77, p = .16, η2 = .04, 95% CI probability of donating in the Moderate Similarity condition [0.00, 0.10]. was higher than the No Information condition, Z=2.64, p = .01, 95% CI [1.49, 14.90]. The differences were marginally statistical significant in two pairs: (1) Low Similarity was Effect of Condition on Decision to Donate marginally higher than the No Information condition, Z= 1.67, p = .09, 95% CI [0.86, 6.76] and (2) High Similarity An omnibus Chi-Square analysis was conducted to examine was marginally lower than the Moderate Similarity condition, the effect of condition on participants’ decision to donate (1 = Z = -1.86, p = .06, 95% CI [0.10, 1.06] (See Table 3). Overall, donated,0=did not donate). This test confirmed that there the results suggested a curvilinear pattern on decision to do- was a statistically significant difference between the highest nate (See Fig. 5). and the lowest conditions, χ (3,140) =8.33, p = .04. The results are presented in order from the highest to the lowest Effect of Condition on Percentage Donated percentage of donors: Moderate Similarity (86.84%), Low Similarity (77.14%), High Similarity (67.74%), No Next, an ANOVA was conducted to examine the effect of Information (58.33%) (See Fig. 5). condition on the percentage of the study payment donated However, a planned pairwise comparison was needed to (see Fig. 6). An omnibus ANOVA result on the full sample, test our specific hypothesis that in the Moderate Similarity including both donors and non-donors, indicated that the condition, participants would be more likely to donate highest average percentage donated was significantly different 5 from the lowest percentage donated, F(3,139) = 3.54, p = .02, η2 = .07, 95% CI [0.002, 0.15]. The average percentage do- 4.5 nated is presented in order: Moderate Similarity: M = 0.30

4 SD = .28, n =38;Low Similarity: M = 0.23, SD = .28, n =35; No Information: M = 0.18, SD = .30, n = 36, and High 3.5 Similarity: M = 0.11, SD = .15, n = 31. The post-hoc pairwise comparisons indicated there were statistically significant dif- 3 ferences in two pairs: (1) participants in the Moderate Self-other similarity Similarity condition donated significantly more of their study 2.5 payment than participants in the No Information condition, 2 t(73) =2.07,p=.04, 95% CI [0.01, 0.25]. Yet (2) participants Low Similarity Moderate Similarity High Similarity in the High Similarity condition donated significantly less of Fig. 4 Average Self-other similarity across conditions the study payment than participants in the Moderate Similarity Curr Psychol

Fig. 5 Percentage of participants 100% d e

donated across conditions t

a 90% n

o 80% D s

t 70% n a

p 60% i c i

t 50% r a

P 40% f o 30% e g a

t 20% n e

c 10% r e

P 0% No Information Low Similarity Moderate High Similarity Similarity condition, t(68) = −3.14, p = .003, 95% CI [−0.32, −0.07]. In organization and cause (Previous Familiarity) were not sig- addition, participants in the High Similarity condition donated nificantly different across conditions. Yet, they may be factors marginally less of the study payment than participants in the that could potentially affect individuals’ charitable donations. Low Similarity condition, t(65) = −1.93, p=.06,95%CI Thus, these three variables were added as covariates in the [−0.25, −0.003]. above analyses to check the robustness of the condition effect The same procedures were used to investigate the sample on the two measures of charitable giving. The results based that only included donors. An omnibus ANOVA test yielded upon a Logit regression (decision to donate as the dependent the same pattern as we obtained from the full sample, but was variable) indicated that both the condition effect and partici- only close to marginally significant, F(3,101) = 2.27, p = .09, pants’ desire to look good were statistically significant predic- η2 = .07, 95% CI [0.00, 0.15]. In the donor only sample, we tors: Z = 2.36, p = .02 (Moderate VS No Information), Z = - found evidence to support the backfiring effect of 1.79, p = .07 (Moderate VS High) and Z = 2.73, p = .01 oversimilarity on percentage of study payment donated: do- (Social Desirability). In addition, the results based upon nors in the High Similarity condition donated statistically and ANCOVAs (for percentage donated) indicated that the effect significantly lower than donors in the Moderate Similarity of condition was still significant on the full sample, F condition, t(68) = −2.56, p = .01, 95% CI [−0.71, −0.96]. In (3,139) = 3.05, p = .03 and close to the marginal significance addition, donors in the High Similarity condition donated mar- on the donor only sample, F (3,101) = 2.15, p=.099, while ginally less than donors in the other two conditions (High none of the three covariates were statistically significant pre- Similarity VS No Information, t(66) = −1.79, p = .08, 95% dictors, ps > .13 (See Table 4). In short, the effect of condition CI [−0.73, −1.02]; and High Similarity VS Low Similarity, did not change after adding the covariates (Prosopagnosia t(65) = −1.80, p = .08, 95% CI [−0.74, −1.02]). Index, Social Desirability Scale, and Previous Familiarity), which indicated that these results were robust.

Robustness Check of the Results Effect of Condition on Liking and Attraction

The randomization check confirmed that participants’ facial An ANOVAwas conducted to examine the effect of condition recognition (Prosopagnosia Index),desiretolookgood on Liking and Attraction. The ANOVA was significant, (Social Desirability Scale), and familiarity with the F(2,103) = 5.72, p =.004, η2 = .10, 95% CI [0.01, 0.21] and

Table 3 Pairwise comparison marginal odds ratios Pairwise comparisons Marginal odds ratios w/o controls

Low Similarity VS No Information 2.41~ Moderate Similarity VS No Information 4.71** High Similarity VS No Information 1.50 Moderate Similarity VS Low Similarity 2.00 High Similarity VS Low Similarity −0.63 High Similarity VS Moderate Similarity −0.32~

~(p < .10), *(p < .05), and **(p <.01) Curr Psychol

Fig. 6 Average percentage 45% d e donated across conditions t a 40% n o

D 35% t n e 30% m y a

P 25% y d

u 20% t S

f 15% o e

g 10% a t n e

c 5% r e

P 0% No Information Low Similarity Moderate High Similarity Similarity All Paritipants Donors Only the results are presented in order from the highest to the low- [−1.60, −0.07], and High VS Moderate Similarity condition, est: Moderate Similarity: M = 4.43, SD = .97, n = 38; High t(68) = −2.46 p = .02, 95% CI [−1.68, −0.18] (See Fig. 8). The Similarity: M = 4.03 SD = 1.00, n = 31; and Low Similarity: findings suggested a backfiring effect of similarity on the M =3.64,SD=1.02, n = 35. Post-hoc pairwise comparisons Beliefs about Donation Impact when similarity became too indicated that participants in the Moderate Similarity high, yet the results did not confirm a positive effect of simi- condition reported higher Liking and Attraction towards the larity when similarity moved from a low to a moderate level. hypothetical donor than participants in the Low Similarity Therefore, there is also no strong supporting evidence for the condition, t(72) = 3.38, p = .001, 95% CI [0.33, 1.26] (See mediating role of Beliefs about Donation Impact in a non- Fig. 7). Yet, there was no difference in Liking and Attraction linear relationship between similarity and donations. between High Similarity and Moderate Similarity condition, t(68) = −1.66, p= .10, 95% CI [−0.88, 0.08] and between High and Low Similarity, t(65) =1.58, p= .12, 95% CI Discussion [−0.10, 0.08]. The findings confirmed a positive effect of similarity on Liking and Attraction when similarity moved from a Although much research has found that similarity to self has a low to a moderate level, yet the results did not provide strong positive influence on a variety of behaviors, including supporting evidence for a backfiring effect of similarity on prosocial behaviors, very little research has suggested that Liking and Attraction when similarity kept increasing. too much similarity could have a negative influence. This Therefore, we are unable to examine the mediating role of experimental study added to this literature by examining Liking and Attraction in a curvilinear relationship between how different levels of similarity to self (low, moderate, high) similarity and donations in this study. among donors could affect charitable giving. With respect to charitable donations, the study found that when other donors Effect of Condition on Beliefs about Donation Impact were moderately similar to the self, participants were more likely to donate to charity (and gave more; See Figs. 5 and An ANOVAwas conducted to examine the effect of condition 6). However, when other donors were high in similarity to the on Beliefs about Donation Impact. The ANOVAwas signifi- self, participants were actually less likely to give (and gave cant based upon the p value but not significant based upon the less) to charity. The results of this study were robust to social 95% confidence interval, F(2,103) = 3.53, p = .03, η2 = .07, desirability, degree of self-reported face-blindness 95% CI [0.00, 0.16], suggesting that at least one pair compar- (prosopagnosia), and previous familiarity with the organiza- ison might be significant. The results are presented in order tion and cause. Taken together, this suggests that there can be from the highest to the lowest impact: Moderate Similarity: such a thing as too much similarity to the self in these types of M =3.74,SD=1.34,n =38;Low Similarity: M =3.64,SD= interactions, and people should not assume that more similar- 1.85, n = 35; and High Similarity: M = 2.81 SD = 1.45, n =31. ity will necessarily lead to more giving in the presence of other Post-hoc pairwise comparisons indicated that the Beliefs generous donors. about Donation Impact in the High Similarity condition was Byrne’s(1961) Similarity Attraction Theory may help to significantly lower than the other two conditions (High VS explain why moving from low to moderate similarity led to Low Similarity condition, t(65) = −2.17, p=.03, 95% CI increased charitable donations in the presence of a generous Curr Psychol

Table 4 Robustness check of the results

Decision to donate (Logit), N = 140 Percentage donated Percentage donated (ANCOVA)-All Participants, N =140 (ANCOVA)-Donors Only, N =102

Condition Moderate condition as the baseline. F(3,139) = 3.05, p =.03 F(3,101) = 2.15, p =.099~ No vs Moderate: Odds ratio = 0.23, SE = .14, Z = -2.36, p = .02* Low vs Moderate: Odds ratio = 0.46, SE = .30, Z = -1.19, p = .23 High vs Moderate: Odds ratio = 0.32, SE = .21, Z = -1.79, p = .07~ Prosopagnosia Index Odds ratio = 1.75, SE = .64, Z = 1.54, p = .124 F(1,139) = 2.30, p =.13 F(1,101) = 0.90, p =.35 Social Desirability Scale Odds ratio = 1.39, SE = .17, Z = 2.73, p =.006** F(1,139) = 0.96, p =.33 F(1,101) = 0.15, p =.70 Previous Familiarity Odds ratio = 0.72, SE = .20, Z = -1.20, p =.23 F(1,139) = 0.04, p =.85 F(1,101) = 0.68, p =.41

All the p values are reported ~(p < .10), *(p <.05),and**(p <.01) donor. The kinship literature may also help to explain why according to the comparisons of group means (See Fig. 7), indi- similarity (especially physical similarity) may influence the cating that too much similarity might cause lower attraction and perception and evaluation of a similar other. The kinship lit- liking compared to similarity at moderate levels, which is contra- erature indicates that individuals can distinguish between dictory to the prediction based upon Similarity Attraction Theory. close genetic kin and non-kin via cue-based mechanisms, thus However, since the condition effect on attraction and liking was behaving differentially towards kin and non-kin (Hamilton marginally significant, more future studies are needed to better 1964). Genetic Similarity Theory has been developed to in- understand whether there is a potential negative effect of similarity corporate the kin selection theory of altruism based upon ge- on attraction and liking, which may lead to lower donations. netic similarity. That is, people are more likely to be altruistic A better explanation for our results is provided by Optimal towards those genetically similar to them (Rushton et al. Distinctiveness Theory (Brewer 1991). Perhaps when individ- 1984). Extending Genetic Similarity Theory, empirical re- uals see that highly similar others have already donated gener- search has found that attitudinal similarity also serves as a ously, they may want to distinguish themselves or their contribu- heuristic cue for kinship recognition, and consequently, peo- tion from others, and thus may donate less (or be less likely to ple tend to behave more prosocially to others who are per- donate). In addition, if individuals believe that they have already ceived to be high in attitudinal similarity (Park and Schaller given via a substitution effect, then there should be no more 2005). Indeed, in our study, we found that participants liked reason to give, and the impact of any donation they make should the other participants more in the moderate similarity condi- be perceived as lower. Indeed, participants in the high similarity tion compared to the low similarity condition. condition rated the impact of their donation as lower than those in However, according to these theories, even more similarity the low and moderate similarity conditions (See Fig. 8). should have an even greater positive effect on donations. In fact, in the current study, donations were less likely (and were smaller) Limitations, and Future Research in the highest similarity condition. Indeed, participants reported liking the other participants less in the high similarity condition To date, the curvilinear relationship between similarity and charitable giving has received limited attention in prior 4.8 4.6 4.5

t 4 c

n 4.4 a o p i 3.5 t m c 4.2 I a n r

t 3 o t i 4 t a A

n 2.5 d o n

3.8 D a

t 2 g u o n i

3.6 b k a 1.5 i s L f

3.4 e i 1 l e

3.2 B 0.5 3 0 Low Similarity Moderate Similarity High Similarity Low Similarity Moderate Similarity High Similarity Fig. 7 Average liking and attraction across conditions Fig. 8 Average beliefs about donation impact across conditions Curr Psychol research. To our knowledge, little evidence has confirmed the explain negative similarity effects. More future research is negative effects at high similarity levels, and no study has yet needed to investigate this possibility. investigated the effects of different levels of similarity on In addition, if we simply divide the gifts into two different prosocial behavior. In order to address this gap in the litera- kinds, tangible ones (i.e. money) and intangible ones (i.e. ture, our study used an innovative approach, facial morphing, volunteering), it is possible for a substitution effect to be happen- to manipulate different levels of similarity to other donors in ing. For instance, people may feel like their responsibility to give order to investigate the potential curvilinear effects of similar- is diminished when someone very close to them gives. This is ity on prosocial behavior. Our study findings indicate that the because tangible gifts can be jointly owned. However, we might facial morphing manipulation works not only in political vot- not expect a similar effect with an intangible gift like giving time. ing contexts, but also in charitable giving contexts. Future For example, because a wife volunteers her time doesn’timply research should also test this effect in other domains of per- that the husband has also donated his time to the same organiza- suasion, such as consumer behavior. Although we manipulate tion, although it might inspire him to be more likely to do so. three levels of similarity, which is rarely done in the literature, In terms of the real world implications of the study find- in our future studies, we will examine even more similarity ings, this study suggests that nonprofits should be careful levels (i.e. 80%, 85%, 90%, 95%) and similarity types (i.e. when they use similarity to encourage donations, because perspective similarity, attitudes similarity, physical similarity too much similarity could potentially backfire in charitable and social similarity, etc.) to help determine the threshold giving. Until more research is conducted, nonprofits should point of too much similarity. aim for moderate similarity (e.g. same gender; see Shang et al. When interpreting our results, readers should be aware that 2007) when encouraging donations, but should avoid high they are based on a situation in which the other donors gave similarity. generously and that they are also based on a sample of college students. Thus, future research is needed to determine if these results would generalize in other contexts, such as when others donate stingily, and in other samples beyond college Conclusion students. Future studies should replicate and extend these results, and try to better understand why they occur. This paper addressed some gaps in the literature by In addition, this study examines the potential for too much positing a curvilinear relationship between donor-donor similarity in the context of donor-donor dyads. But future similarity and one kind of prosocial behavior, charitable studies are needed in order to examine whether this theory is giving. Research finds positive relationships between generalizable to other charitable giving contexts, such as similarity and a variety of behaviors. Yet, comparatively donor-recipient or donor-solicitor dyads. Since research has little is known about whether individuals could respond also indicated a positive effect of similarity on other behaviors negatively to others’ generous donations when they are such as compliance, consumer behaviors, aggression, and dat- too similar to the self. What we do know, based upon ing (See Introduction), future research should also go beyond the results of this paper, is that there is not a simple the charitable giving context to see whether there is a such linear and positive relationship, but rather a curvilinear thing as too much similarity in these domains as well. We relationship, between self-other similarity and charitable think that over-similarity may be a general principle that giving. That is, too much similarity between donors can would likely apply to a variety of contexts, but future research sometimes backfire when others give generously. will help to determine this. Additionally, we intended to examine the underlying pro- ———————————— cess of the similarity effect on donation behavior by investigating the relevant measures, such as Liking, Attraction and Compliance with Ethical Standards Donors’ Beliefs of Donation Impact in our study, because the literature suggested that those might be the mediator factors. Conflict of Interest On behalf of all authors, the corresponding author However, we did not find strong supporting evidence that states that there is no conflict of interest. these factors explained the curvilinear relationship between Research Involving Human Participants All procedures performed in similarity and donation behavior in our study. Based upon studies involving human participants were in accordance with the ethical these unexpected findings, it is still unclear what caused the standards of the institutional and/or national research committee and with curvilinear effect of similarity on donations. It is possible that the 1964 Helsinki declaration and its later amendments or comparable the positive and negative effects of similarity on donation ethical standards. could be explained by two different mechanisms. For exam- Informed Consent Informed consent was obtained from all individual ple, Liking and Attraction might explain positive similarity participants included in the study. effects, and Donor’s Beliefs about Donation Impact might Curr Psychol

Data Statement The datasets generated during and/or analyzed during Kelman, H. C. (1958). Compliance, identification, and internalization the current study are not publicly available due the funding restriction but three processes of attitude change. Journal of Conflict Resolution, are available from the corresponding author on reasonable request. 2(1), 51–60. Konrath, S., & Handy, F. (2018). The development and validation of the motives to donate scale. Nonprofit and Voluntary Sector Quarterly, 47(2), 347–375. References Konrath, S., Bushman, B. J., & Campbell, W. K. (2006). Attenuating the link between threatened egotism and aggression. Psychological – Andsager, J. L., Bemker, V., Choi, H.-L., & Torwel, V. (2006). Perceived Science, 17(11), 995 1001. similarity of exemplar traits and behavior: Effects on message eval- Krupp, D. B., Debruine, L. M., & Barclay, P. (2008). A cue of kinship uation. Communication Research, 33(1), 3–18. promotes cooperation for the public good. Evolution and Human – Aron, A., Aron, E. N., Tudor, M., & Nelson, G. (1991). Close relation- Behavior, 29(1), 49 55. ships as including other in the self. Journal of Personality and Social Lee, L., Piliavin, J. A., & Call, V. R. (1999). Giving time, money, and Psychology, 60(2), 241. blood: Similarities and differences. Social Psychology Quarterly, – Bailenson, J. N., Iyengar, S., Yee, N., & Collins, N. A. (2009). Facial 276 290. similarity between voters and candidates causes influence. Public Lott, A. J., & Lott, B. E. (1965). Group cohesiveness as interpersonal Opinion Quarterly, 72(5), 935–961. attraction: A review of relationships with antecedent and consequent – Bakar, H. A., & McCann, R. M. (2014). Matters of demographic similar- variables. Psychological Bulletin, 64(4), 259 309. ity and dissimilarity in supervisor–subordinate relationships and Mitteness, C. R., DeJordy, R., Ahuja, M. K., & Sudek, R. (2016). workplace attitudes. International Journal of Intercultural Extending the role of similarity attraction in friendship and advice Relations, 41,1–16. networks in angel groups. Entrepreneurship Theory and Practice, 40(3), 627–655. Baron, R. A. (1971). Aggression as a function of magnitude of victim's pain cues, level of prior anger arousal, and aggressor-victim similar- Moreland, R. L., & Zajonc, R. B. (1982). Exposure effects in person ity. Journal of Personality and Social Psychology, 18(1), 48–54. perception: Familiarity, similarity, and attraction. Journal of Experimental Social Psychology, 18(5), 395–415. Bekkers, R. (2010). George gives to geology Jane: The name letter effect Park, J. H., & Schaller, M. (2005). Does attitude similarity serve as a and incidental similarity cues in fundraising. International Journal heuristic cue for kinship? Evidence of an implicit cognitive associ- of Nonprofit and Voluntary Sector Marketing, 15(2), 172–180. ation. Evolution and Human Behavior, 26(2), 158–170. Ben-Ner, A., McCall, B. P., Stephane, M., & Wang, H. (2009). Identity Penton-Voak, I., Perrett, D., & Peirce, J. (1999). Computer graphic studies and in-group/out-group differentiation in work and giving behav- of the role of facial similarity in judgements of attractiveness. iors: Experimental evidence. JournalofEconomicBehavior& Current Psychology, 18(1), 104–117. Organization, 72(1), 153–170. Rushton, J. P., Russell, R. J., & Wells, P. A. (1984). Genetic similarity Boivin, M., Dodge, K. A., & Coie, J. D. (1995). Individual-group behav- theory: Beyond kin selection. Behavior Genetics, 14(3), 179–193. ioral similarity and peer status in experimental play groups of boys: Secord, P. F., & Backman, C. W. (1964). Interpersonal congruency, per- The social misfit revisited. Journal of Personality and Social ceived similarity, and friendship. Sociometry,115-127. Psychology, 69(2), 269–279. Shah, P., Gaule, A., Sowden, S., Bird, G., & Cook, R. (2015). The 20- Brewer M.B. (2016) The Social Self: On Being the Same and Different at item prosopagnosia index (PI20): A self-report instrument for iden- the Same Time. Personality and Social Psychology Bulletin, 17(5), tifying developmental prosopagnosia. Royal Society Open Science, 475–482. 2(6), 140343. Burger, J. M., Messian, N., Patel, S., del Prado, A., & Anderson, C. Shang, J., Croson, R., & Reed, A. (2007). “I” give, but “we” give more: (2004). What a coincidence! The effects of incidental similarity on The impact of identity and the mere social information effect on compliance. Personality and Social Psychology Bulletin, 30(1), 35– donation behavior. NA-Advances in Consumer Research Volume, 43. 34. Byrne, D. (1961). Interpersonal attraction and attitude similarity. The Silvia, P. J. (2005). Deflecting reactance: The role of similarity in increas- Journal of Abnormal and Social Psychology, 62(3), 713. ing compliance and reducing resistance. Basic and Applied Social Cialdini, R. B., & Goldstein, N. J. (2004). Social influence: Compliance Psychology, 27(3), 277–284. – and conformity. Annual Review of Psychology, 55,591 621. Sinervo, B., & Clobert, J. (2003). Morphs, dispersal behavior, genetic Eagly, A. H., & Koenig, A. M. (2006). Social Role Theory of Sex similarity, and the evolution of cooperation. Science, 300(5627), Differences and Similarities: Implication for Prosocial Behavior: 1949–1951. Lawrence Erlbaum Associates Publishers. Sole, K., Marton, J., & Hornstein, H. A. (1975). Opinion similarity and Hamilton, W. D. (1964). The genetical evolution of social behaviour. II. helping: Three field experiments investigating the bases of promo- Journal of Theoretical Biology, 7(1), 17–52. tive tension. Journal of Experimental Social Psychology, 11(1), 1– Heldt, T. (2005). Conditional cooperation in the field: Cross-country 13. skiers’ behavior in Sweden. Department of Economics and Strahan, R., & Gerbasi, K. C. (1972). Short, homogeneous versions of the Society, Dalarna University, Sweden. Marlowe-Crowne social desirability scale. Journal of Clinical Helms, S. E., & Thornton, J. P. (2012). The influence of religiosity on Psychology. charitable behavior: A COPPS investigation. The Journal of Socio- Tian, Y., & Konrath, S. (2019). The effects of similarity on charitable Economics, 41(4), 373–383. giving in donor–donor dyads: A systematic literature review. Hendrick, C., & Taylor, S. P. (1971). Effects of belief similarity and Voluntas: International Journal of Voluntary and Nonprofit aggression on attraction and counteraggression. Journal of Organizations, 1-24. (Online First). Personality and Social Psychology, 17(3), 342–349. USA, G. (2019). The Annual Report on Philanthropy for the Year, 2018. Jetten, J., Spears, R., & Manstead, A. (1998). Intergroup similarity and Van Hoye, G., & Turban, D. B. (2015). Applicant–employee fit in per- group variability: The effects of group distinctiveness on the expres- sonality: Testing predictions from similarity-attraction theory and sion of ingroup bias. Journal of Personality and Social Psychology, trait activation theory. International Journal of Selection and 74(6), 1481–1492. Assessment, 23(3), 210–223. Curr Psychol

Woodside, A. G., & Davenport, J. W. (1974). The effect of salesman Publisher’sNoteSpringer Nature remains neutral with regard to jurisdic- similarity and expertise on consumer purchasing behavior. Journal tional claims in published maps and institutional affiliations. of Marketing Research, 11(2), 198–202. Yinon, Y., & Sharon, I. (1985). Similarity in religiousness of the solicitor, the potential helper, and the recipient as determinants of donating behavior. Journal of Applied Social Psychology, 15(8), 726–734.